MIVIA-Crime: A Benchmark for Cross-Dataset Evaluation of Video Anomaly Detection

MIVIA-Crime is a surveillance video benchmark designed for cross-dataset evaluation of video anomaly detection methods. The dataset provides temporally annotated anomalous events together with anomaly class labels, enabling the evaluation of temporal localization and online anomaly detection algorithms under distribution shift. The complete dataset, documentation, and evaluation scripts will be released after acceptance of the associated paper Online Fully Supervised Video Anomaly Detection.